joergdietrich / daltonize Goto Github PK

Fixing the invariant to red for tritanopia (#15 ) seems to break the correction transformation for daltonization of tritanopia.

I know this is an old repo but this is still being used by several developers (including me).

I noticed that on a 2GB linux machine , MemoryError is raised when converting a 2-3MB image (3933 x 5906 pixels in size). Here is a truncated error message.

  File "./<app>/daltonize.py", line 132, in simulate_from_image
    rgb = np.asarray(img, dtype=float)
  File "/root/.cache/pypoetry/virtualenvs/<path>/lib/python3.6/site-packages/numpy/core/_asarray.py", line 85, in asarray
    return array(a, dtype, copy=False, order=order)
MemoryError: Unable to allocate 532. MiB for an array with shape (3933, 5906, 3) and data type float64

This is very weird that for a 3MB file, it requires a large 532MB amount of memory.

Related line on this repo: https://github.com/joergdietrich/daltonize/blob/master/daltonize.py#L79

The RGB-LMS conversion matrices and color blindness matrices used by this repo are very different from those listed on https://ixora.io/projects/colorblindness/color-blindness-simulation-research/ (which you mentioned you consulted in #13), and also those in the original MATLAB file in https://github.com/joergdietrich/daltonize/blob/master/doc/conv_img.m.

I'm quite confused why these are all different. Could you maybe shed some light on this?

On my Ubuntu 20.04 with Python 3.8.10 and numpy 1.21.3 I always get black images out of the simulation function (all values are zero). numpy 1.20.3 works fine, the problem started with numpy 1.21.0.

The guilty part is transform_colorspace, where the np.einsum("ij, ...j", mat, img, dtype=np.float16, casting="same_kind") always returns 0 when given rgb2lms as input, unless dtype is changed back to np.float32. As a consequence this also breaks the daltonization, which becomes a noop.

I'm not sure if this is actually a numpy bug, but an alternative solution is to return img @ mat.T, which is equivalent to the einsum thanks to numpy broadcasting rules, but still works with float16. It may also be a bit faster.

Going back to float32 might be a safer choice too!

I get the following error when I run the program on a 800x600 PNG image:

ValueError: operand 0 did not have enough dimensions to match the broadcasting, and couldn't be extended because einstein sum subscripts were specified at both the start and end

Hello,

This script gives me different results from other color blindness simulator.

Steps to reproduce:

1. Select an image
2. Simulate color blindness with t, d or p
3. Compare the results with https://www.color-blindness.com/coblis-color-blindness-simulator/ or https://logolab.app/lab

Expectation:
The results are the same

Actual:
The results are very different

Data image used:

The simulation function for color vision deficiency doesn't work for the current matplotlib version (3.5.1).